Comment on Three X-ray Crystal Structure Papers

Experiences From Developing Software for Large X-Ray Crystallography-Driven Protein-Ligand Studies

The throughput of macromolecular X-ray crystallography experiments has surged over the last decade. This remarkable gain in efficiency has been facilitated by increases in the availability of high-intensity X-ray beams, (ultra)fast detectors and high degrees of automation. These developments have in turn spurred the development of several dedicated centers for crystal-based fragment screening which enable the preparation and collection of hundreds of single-crystal diffraction datasets per day. Crystal structures of target proteins in complex with small-molecule ligands are of immense importance for structure-based drug design (SBDD) and their rapid turnover is a prerequisite for accelerated development cycles. While the experimental part of the process is well defined and has by now been established at several synchrotron sites, it is noticeable that software and algorithmic aspects have received far less attention, as well as the implications of new methodologies on established paradigms for structure determination, analysis, and visualization. We will review three key areas of development of large-scale protein-ligand studies. First, we will look into new software developments for batch data processing, followed by a discussion of the methodological changes in the analysis, modeling, refinement and deposition of structures for SBDD, and the changes in mindset that these new methods require, both on the side of depositors and users of macromolecular models. Finally, we will highlight key new developments for the presentation and analysis of the collections of structures that these experiments produce, and provide an outlook for future developments.

Detect, correct, retract: How to manage incorrect structural models

The massive technical and computational progress of biomolecular crystallography has generated some adverse side effects. Most crystal structure models, produced by crystallographers or well-trained structural biologists, constitute useful sources of information, but occasional extreme outliers remind us that the process of structure determination is not fail-safe. The occurrence of severe errors or gross misinterpretations raises fundamental questions: Why do such aberrations emerge in the first place? How did they evade the sophisticated validation procedures which often produce clear and dire warnings, and why were severe errors not noticed by the depositors themselves, their supervisors, referees and editors? Once detected, what can be done to either correct, improve or eliminate such models? How do incorrect models affect the underlying claims or biomedical hypotheses they were intended, but failed, to support? What is the long-range effect of the propagation of such errors? And finally, what mechanisms can be envisioned to restore the validity of the scientific record and, if necessary, retract publications that are clearly invalidated by the lack of experimental evidence? We suggest that cognitive bias and flawed epistemology are likely at the root of the problem. By using examples from the published literature and from public repositories such as the Protein Data Bank, we provide case summaries to guide correction or improvement of structural models. When strong claims are unsustainable because of a deficient crystallographic model, removal of such a model and even retraction of the affected publication are necessary to restore the integrity of the scientific record.

A multi-crystal method for extracting obscured crystallographic states from conventionally uninterpretable electron density

In macromolecular crystallography, the rigorous detection of changed states (for example, ligand binding) is difficult unless signal is strong. Ambiguous (‘weak' or ‘noisy') density is experimentally common, since molecular states are generally only fractionally present in the crystal. Existing methodologies focus on generating maximally accurate maps whereby minor states become discernible; in practice, such map interpretation is disappointingly subjective, time-consuming and methodologically unsound. Here we report the PanDDA method, which automatically reveals clear electron density for the changed state—even from inaccurate maps—by subtracting a proportion of the confounding ‘ground state'; changed states are objectively identified from statistical analysis of density distributions. The method is completely general, implying new best practice for all changed-state studies, including the routine collection of multiple ground-state crystals. More generally, these results demonstrate: the incompleteness of atomic models; that single data sets contain insufficient information to model them fully; and that accuracy requires further map-deconvolution approaches.

Building a ligand into a weak region of an electron density map of a protein is a subjective process. Here, the authors present a new method to obtain a clear electron density for a bound ligand based on multi-crystal experiments and 3D background correction.

Twilight reloaded: the peptide experience

The de facto commoditization of biomolecular crystallography as a result of almost disruptive instrumentation automation and continuing improvement of software allows any sensibly trained structural biologist to conduct crystallographic studies of biomolecules with reasonably valid outcomes: that is, models based on properly interpreted electron density. Robust validation has led to major mistakes in the protein part of structure models becoming rare, but some depositions of protein-peptide complex structure models, which generally carry significant interest to the scientific community, still contain erroneous models of the bound peptide ligand. Here, the protein small-molecule ligand validation tool Twilight is updated to include peptide ligands. (i) The primary technical reasons and potential human factors leading to problems in ligand structure models are presented; (ii) a new method used to score peptide-ligand models is presented; (iii) a few instructive and specific examples, including an electron-density-based analysis of peptide-ligand structures that do not contain any ligands, are discussed in detail; (iv) means to avoid such mistakes and the implications for database integrity are discussed and (v) some suggestions as to how journal editors could help to expunge errors from the Protein Data Bank are provided.

Correcting the record of structural publications requires joint effort of the community and journal editors

Seriously flawed and even fictional models of biomolecular crystal structures, although rare, still persist in the record of structural repositories and databases. The ensuing problems of database contamination and persistence of publications based on incorrect structure models must be effectively addressed. The burden cannot be simply left to the critical voices who take the effort to contribute dissenting comments that are mostly ignored. The entire structural biology community, and particularly the journal editors who exercise significant power in this respect, must engage in a constructive dialog lest structural biology lose its credibility as an evidence-based empirical science.